~/onflow.studio/ci-flow

CI Flow

Run GitHub Actions on a spare machine and watch it from anywhere.

The laptop in the drawer becomes your CI. Open source, MIT.

Set up a machine Star on GitHub

cimon: the machine, its runners, what's running and what just finished, on one screen.

# why

CI minutes run out. Old laptops don't.

A spare machine is warm and already paid for. CI Flow gives it clean runners, a dashboard and no open ports. No Kubernetes, no cloud account.

# what you get

  • 01

    Your hardware, your minutes

    No per-minute bill, warm caches. GitHub stays your CI.

  • 02

    A fresh runner per job

    Its own container and Docker daemon, cleaned when the job ends.

  • 03

    cimon, in the terminal

    Machine health, runners, pipelines and results, live.

  • 04

    The same view on your phone

    A web page on your tailnet. No login to set up.

  • 05

    Alerts when it matters

    Runner down, disk full, too hot: a push to your phone.

  • 06

    Nothing public

    The machine only talks out to GitHub.

# on your tailnet

Check CI from your phone.

Same view as cimon, live, shared only on your tailnet.

The CI Flow web dashboard: machine health, six of seven runners busy, three running pipelines with their jobs, one run queued and the recent results for three repositories
The web dashboard, with fictional repositories.

# security

A runner runs what the workflow says.

On a public repository, a fork's pull request can change it. Before you add one:

  1. 1.Require approval for fork pull requests.
  2. 2.Keep forks off the runner with one if: line.
  3. 3.ci-flow warns you when a repository is risky.

# install

One command plans it all.

On the spare machine. Without --yes it only prints the plan. You'll need:

  • A spare Linux machine with sudo
  • Tailscale, and gh logged in
  • Python 3

Then, in the repository's workflows: runs-on: [self-hosted, ci-box].

$ git clone https://github.com/onflow-studio/ci-flow$ cd ci-flow$ ./ci-flow init  # the plan$ ./ci-flow init --yes  # do it$ ci-flow add you/repo --runners 2 --labels ci-box$ ci-flow apply

On a fresh machine, init:

  1. 1.Installs Docker, Tailscale and gh if missing.
  2. 2.Keeps the runners up with nobody logged in.
  3. 3.Stops a closed laptop lid from suspending it.
  4. 4.Cleans old images every week.

# stack

ci-flow --stack

lang
Python, standard library only
runs
systemd · Docker · GitHub's runner
reach
your tailnet only
tests
run on ci-flow itself

# how it works

Each runner slot is a Docker daemon plus a fresh runner container per job. When the job ends, the slot is wiped.

One collector watches the machine and GitHub. cimon and the web page both read it.

# origin

Hand-made runners on a spare laptop, kept busy by Shogun Flow's agents. The copies drifted, so they became one tool. Its own tests run on it.

# make it yours

Fork it, make it yours.

Don't adapt to it: fork it and have your AI make it work your way. Next up: several machines sharing the load.

Set up a machine Star on GitHub